Evaluation and Comparison of the Artificial Neural Network and the HEC-HMS Models in the Simulation of the Rainfall-Runoff process and the development of Hydrograph in the Kasilian representative Basin

The Rainfall-runoff process is a non-linear and a very complicated phenomenon. As collection of the reliable data is difficult, time consuming and expensive, hydrologists resort to the simulation of such events using the so-called black box models such as the artificial neural network (ANN). However, as such models have been developed and evaluated and different geographical settings, their comparison is essential if one desires to apply them to a certain watershed. To this end, the ANN (version 9-10-7) and the HEC-HMS models were evaluated and copmarid in generating hydroghraph for the kasilian basin, to improve the models stability and training, the rainfall data were divided into four groups according to the Huff distribution of rainfall pattern. Furthermore, different combinations of transfer functions were used in the hidden and output layers. The ANN model was derived using the Qnet2000 software. The HEC-HMS model was also used to compare it with the ANN.
the absolute relative error of QP, TP, Tb, W75, W50, T50 and T75 parameters simulated using the ANN were 0.02-51.97, 0.55-41.23, 0.26-54.07, 0.23-202.62, 0.52-69.88, 2.21-82.07 and 2.42-55.76, respectively. Meanwhile these confines were 0.58-756.53, 0-250, 0-141.18, 2.84-575, 0.93-167.86, 3.33-350 and 2-266.67 using the HEC-HMS model. Regarding the relative error of the outcomes of each event, it can be concluded that the neural network in the most cases has been simulated all the parameters and the overall shape of the hydrograph with little error compared to the HEC-HMS model. Ofcourse the HEC-HMS model was rarely more accurately than the ANN in the some cases, for example, to simulate the peak, the base time and overall shape of hydrograph.